We propose a novel method for HDR image reconstruction from a single LDR image based on fully end-to-end deep CNN architecture that uses residual learning and attention mechanism. Our network can restore the dynamic range and recover information lost in over-exposed areas with the help of a contextual attention module. Quantitative and qualitative experiments show competitive results of the proposed method compared to previous approaches. Furthermore, an ablation study results show the ability of the residual learning to boost performance and demonstrate the significance of using the attention mechanism.